Loading a Max for Live patch from disk into Ableton blocks the main process

I've got a Max for Live patch for which I'm trying to optimize the loading time or stop it from blocking the Ableton process. When loading the patch in Ableton on Windows 10/11 it can take up to 30 seconds and blocks the main Ableton process during this time (screen is frozen). On MacOS, same Ableton version, it also blocks the main Ableton process but only for a few seconds which makes sense, this is also acceptable.

I've done some monitoring on the Ableton process statistics when loading the device and already identified that reading the device data from disk into Ableton is a huge bottleneck. The device is definitely not small (~50MB due to embedded ML models) but monitoring shows an average data transfer rate of 10MB/s. This doesn't really add up because the actual load time is 30 seconds. Step one is to reduce the device size as much as possible.

Besides shrinking the models / assets included, is there any way to "clean" the device before / while exporting to Max for Live?

This also still doesn't answer my question why this process takes so much longer on Windows than on MacOS. Does someone know more about this and can explain what's happening in the background when a Max for Live device is loaded into Ableton. Thanks!

---
Max 8.3.0
Ableton 10.1.42