More parameters needs to be defined for a realistic response to this question.
The real-life feasibility of a mecha design is already beaten like a deadhorse, but you are specifically rejecting those answers, so there isn't a whole lot to work with here. From your question and some of your comments it seems like you are under the imrpession mecha designs are not practical only due to bipedal design; this is not true, and if anything it is quite misguided. Mechwarrior style bidpedal/quadripedal weapon platforms are more practical than a Wall-E mecha since it can perform its designed functional task. The multi-legged design doesn't intrinsictly hinder that function, and might be justified by the terrain it is intended to operate in, such as over crevaces and uneven terrain.
Mecha as a concept is impractical. The closest analog would be the exosuit as someone else has posted.
Every machine is designed with a functional purpose to carry out tasks within the paremeters defined for it. Does it need to resist heat? Ok it might use heat-resistant alloys. Does it need to withstand sudden rockslides? Okay, it might have thick plates. If the functional task is to lift heavy objects, we have cranes. If the task is to deploy weapons, we have tanks and airplanes.
If you aim is for a multirole-everything - it would not be cost effective and may not ever be technologically possible at gundam-sizes. If you can be satisfied with human-sizes, those are called androids.
A real-life large sized mecha fight already happened: Megabot vs Suidobashi. It was... boring and underwhelming. Even witht he most advanced and powerful hydraulics system to lift the arms, any weapon platform (a jet or tank, even a human with a bazooka) can destroy the mecha before can point forward.
To answer your question there must be a clear situation and problem to solve to justify the mecha's design. In my world it is a densely packed jungle zone where tank treads are impractical. A rogue-AI machine colony was destroyed and AI outlawed but the humans repurposed their multirole logging/worker mechas by ripping out the core and rebuild it into a cockpit. A scenario like mine wouldn't really have a specific "upper limits for design" to be relevant.