If the 25% chance is based on population percent with the gene...
If the magic gene offers an advantage, the percentage will raise over time. If the magic gene offers a disadvantage, it will decrease. If the magic gene offers a net balance, nothing will necessarily happen either way; other variables will determine the course history takes.
...Oh? But the two segments of society are separated?
Let's say there are 1 million people. Exactly 25% (250,000) are Mages. 75% (750,000) are Mundanes. Let's use a Population Calculator to see what the percentage will look like in 100 years, assuming a 1.2% growth rate for both.
Mages: 824,121 (24.999992416%)
Mundanes: 2,472,364 (75.000007583%)
TOTAL: 3,296,485 (100%)
As you can see, unless an outside force affects that growth percentage, it will maintain 25/75 split. (I even tested for 1000 years later and it was still the same, I'm just sparing you the monstrous numbers.)
That said, this is a 25% chance a person will be born into the magical society, not a 25% chance that they will be born with the Mage genes. There's a clear distinction, but this does allow for the first to hold true... mostly. There's another caveat I didn't include: This assumes that the population of Mundanes don't have the Mage gene at all. If they do have the Mage gene, then the odds are going to be different based on who is breeding with who, as you can see in the next section.
If the 25% chance is based on genetics...
This only holds true if a mundane with the recessive magic gene breeds with another mundane with the recessive magic gene.
As Mundanes breed together, there's a 50% chance that a given Mundane has the gene and a 50% chance they don't. There's a 25% chance that both Mundanes have the gene. If one has the gene, but the other doesn't, there's a 100% chance the child will be born Mundane. If both have the gene, there's a 75% chance the child will be born mundane and 25% chance the child will be born a mage. So, let's assume that two random Mundanes breed, assuming a truly equal distribution of Mundanes with and without the Mage gene. There would be a 6.25% chance that two Mundanes would have a Mage child.
But Sora, where are you getting 6.25% from?
When you're calculating the odds of this event happening, you have to calculate it by looking at the likelihood of the events happening separately and then combining them. Let me clarify. There's a 50% chance that a given Mundane will have the gene. Since we need two Mundanes, that means we would consider the likelihood twice: 50% of 50% is 25%. So, with that, we now know there's a 25% chance that two Mundanes that both have the recessive gene will breed.
If you look at the following Punnett Square, you can see the odds of a Mage (mm) being born from 2 Mundanes with the Mage gene is 1 out of 4 or 25%.

Since we're trying to see the odds of getting a mage from 2 random Mundanes, we know the odds of 2 Mundanes both having the right gene is 25% and now we know that the odds of a Mage being born from them is also 25%. So, we have to find out what 25% of 25% is: 1/16 or 6.25%.
If two Mages breed together or a Mundane without the recessive gene breeds with anyone, (Mage, Mundane, it doesn't matter,) you're guaranteed 100% outcome of Mage and 100% outcome of a Mundane, respectively.
below: The odds of getting a Mage from 2 Mages... 100%

below: The odds of getting a Mage from a Mage and a Mundane without the Mage gene... 0%

above: The odds of getting a Mage from a Mundane with and a Mundane without the Mage gene... 0%

So, let's say we want to see what the odds are of a Mage being born under the following assumptions:
- Equal Mage and Mundane Population
- Equal chance a Mundane does or does NOT have the Mage gene
- Mage gene is recessive
- Societies are NOT segregated
- There are no social barriers barring relationships
- There is no genetic reason to prefer Mundane or Mage over the other
- Our sample is all equally likely to be involved with someone else from the group
Well to test the odds of this, we have 4 males and 4 females. 2 males and 2 females are Mages. 1 male and 1 female are Mundanes with the Mage gene. 1 male and 1 female are Mundanes without the Mage gene. The outcomes we are testing for will be as follows:
- Mundane born
- a) Mundane with Mage gene born
- b) Mundane without Mage gene born
- Mage born
First, let's check the odds of an individual male winding up with an individual female... (or vice versa. The order here is irrelevant.) That means each male has a 25% chance of picking any given female (assuming no conflicts and fighting over choice of mate occur).
There's a 50% chance the male is a Mage and a 50% he's not a Mage.
Let's assume we grab one of the 2 male Mages. There's a 50% chance his mate will be a Mage and 50% chance she won't be a Mage. That brings the odds of a Mage being born to become 25%. We've been over this already.
But what about that 50% chance his mate is NOT a Mage? As we saw, if the female Mundane does NOT have the Mage gene, their children will NOT be born mages, but they will all have the Mage gene. If the female Mundane does have the Mage gene, we get something different.

As you can see, there is now a 50% chance that the child will be born as a Mage or Mundane. I feel like you get the general idea, so instead of repeating the same song and dance, let's look at these percents so far:
Odds Checker
A: Pairing - Odds of Each Pairing
(Who selects who is irrelevant. For simplicitity I am using "Male X selects", but it could just as easily be "Female X selects".)
- Male Mage (50%) selects Female Mage (50%) - 25%
- Male Mage (50%) selects Female Mundane with Mage Gene (25%) - 12.5%
- Male Mage (50%) selects Female Mundane without Mage Gene (25%) - 12.5%
- Male Mundane with Mage Gene (25%) selects Female Mage (50%) - 12.5%
- Male Mundane with Mage Gene (25%) selects Female Mundane with Mage Gene (25%) - 6.25%
- Male Mundane with Mage Gene (25%) selects Female Mundane W/O Mage Gene (25%) - 6.25%
- Male Mundane W/O Mage Gene (25%) selects Female Mage (50%) - 12.5%
- Male Mundane W/O Mage Gene (25%) selects Female Mundane with Mage Gene (25%) - 6.25%
- Male Mundane W/O Mage Gene (25%) selects Female Mundane W/O Mage Gene (25%) - 6.25%
B - Results of Each Possible Pairing
*(This is going to assume 4 children from each possible pair. Additionally, this will use Punnett Square notation. M - Mundane gene; m - Mage gene)
- Male Mage 1 (mm) x Female Mage 1 (mm) - mm, mm, mm, mm (4 Mages)
- Male Mage 1 (mm) x Female Mage 2 (mm) - mm, mm, mm, mm (4 Mages)
- Male Mage 1 (mm) x Female Mundane W/ Gene (Mm) - Mm, Mm, mm, mm (2 Mages, 2 W/)
- Male Mage 1 (mm) x Female Mundane W/O Gene (MM) - Mm, Mm, Mm, Mm (4 W/)
- Male Mage 2 (mm) x Female Mage 1 (mm) - mm, mm, mm, mm (4 Mages)
- Male Mage 2 (mm) x Female Mage 2 (mm) - mm, mm, mm, mm (4 Mages)
- Male Mage 2 (mm) x Female Mundane W/ Gene (Mm) - Mm, Mm, mm, mm (2 Mages, 2 W/)
- Male Mage 2 (mm) x Female Mundane W/O Gene (MM) - Mm, Mm, Mm, Mm (4 W/)
- Male Mundane W/ Gene (Mm) x Female Mage 1 (mm) - Mm, Mm, mm, mm (2 Mages, 2 W/)
- Male Mundane W/ Gene (Mm) x Female Mage 2 (mm) - Mm, Mm, mm, mm (2 Mages, 2 W/)
- Male Mundane W/ Gene (Mm) x Female Mundane W/ Gene (Mm) - MM, Mm, mM, mm (1 Mage, 2 W/, 1 W/O)
- Male Mundane W/ Gene (Mm) x Female Mundane W/O Gene (MM) - MM, MM, Mm, Mm (2 W/, 2 W/O)
- Male Mundane W/O Gene (MM) x Female Mage 1 (mm) - Mm, Mm, Mm, Mm (4 W/)
- Male Mundane W/O Gene (MM) x Female Mage 2 (mm) - Mm, Mm, Mm, Mm (4 W/)
- Male Mundane W/O Gene (MM) x Female Mundane W/ Gene (Mm) - MM, MM, Mm, Mm (2 W/, 2 W/O)
- Male Mundane W/O Gene (MM) x Female Mundane W/O Gene (MM) - MM, MM, MM, MM (4 W/O)
Total Children born assuming 4 from every pairing:
- 25 Mages
- 30 Mundanes With the Gene
- 9 Mundanes Without the Gene
TOTAL: 64
Odds of a Mage being born - 39.0625%
Odds of a Mundane With the Gene being born - 46.8750%
Odds of a Mundane Without the Gene being born - 14.0625%
B shows that assuming our sample (Generation 1) resulting in a 50/50 split between Mages and Mundanes AND assuming they all have 4 children with each possible mate, you will get a decrease in the percentage of Mages, in Generation 2. In Generation 3, we see that, doing the same process results in the same percentage and ratio (as well as birth defects from in-breeding [/AYO!]).
Of course, all of this is assuming specific circumstances. In reality, probability can't be assumed to go so clean-cut and clear. Environmental causes may very well cause Mages to swell or cull in numbers. You can't assume the statistic of what should happen in a controlled environment will always reflect what path reality takes.