Sorry for the late reply. It was a long day. Took awhile to get back at it.
Thank you both, RainyShadow and Jaclaz.
Rainy, you have a good point. I had not thought of that. But I do not have Ps2 connectors. Maybe if I order a serial/com mouse it would install that before the USB. Its been years since I've even owned a com mouse. I don't know how well 95 picks that up, during boot.
Jaclaz, the booting over USB is done by using a fat32 partition with msdos io.sys and mbr. So the bootloader is dos. Grub4dos is loading an 80mb image into ram and booting that (win95 with ultrapack drivespace3).
I'll confess some things not originally mentioned. I left them out, as to not confuse the situation. The device does have IDE and I can boot dos from that. However, as soon as any advanced OS attempts to boot from the IDE (the exception KolibiOS), the connection to IDE is lost. This is intentionally set in the bios by the manufacturer. Essentially the bios is marked as disabled. But it is a little more complex than that, and the details are in-depth.
So I have syslinux chainload grub4dos or grub2. It is from the IDE that grub4dos loads the 80mb.img to ram. But if Win95 can in anyway see any partition, on the IDE, boot will fail. It does not matter what boot switches you use. And it does not matter if you are not even using the IDE. At some point Windows cannot access its own files and errors out, in some way or another. Or if booting from ram, and the partition is still visible, you end up with a protection error. The same is true trying safe mode, either way. That is why some very ambitious people wrote a coreboot bios replacement for the device. So the Win95 image needs to be in ram, and it needs to be mapped over the any IDE partition(s)[or mark them hidden]. I do believe that, eventually, I may find a way to get IDE access working. But only after working from inside Windows. And it will be a terrible hack, I am sure.
Linux can boot from ram, but cannot see the IDE. You can edit the (Linux) ATA driver to use MSR configurations instead of PCI, and activate the MSRs related to IDE enabling. But using grub2 with the same MSRs modified (wrmsr) changes nothing for Win95. Windows uses the PCI configurations, so the MSR modifications hardly mean anything.
Booting anything from USB can be rough. Some people have gotten certain sizes and manufacturers' pendrives to boot. While other do not boot at all.
The USB booting issues arrive from the bootloaders hardly working at all, with this device. For me, dos boot over USB works every time; no matter what. In a few minutes I will try direct mapping the image from USB, booting from USB dos mbr, and directly running grub4dos (grub.exe) from the prompt. I may have tried that before, but actually can't remember for certain. The only problem is that USB is 1.1 speeds. For now that would work. I'm hoping this way just slipped my mind. Or that when I, possibly, tried it before, I hadn't learned that Windows would boot; if it can't access an IDE partition. I've been working on this for a long time. Masochist, I am.
Anyway, this is a little bit more of a full disclosure.
In the end, I intend to run from ram anyway. Drivespace is actually pretty cool, now. Back when 95 was new, I wouldn't have touched it with a ten foot pole. With faster memory and CPU, its kinda like Squashfs for Linux. If only there was something like Aufs for Win9x. Trying to set this mean device up, has taught me some neat things. For example, I learned how to modify Grub2 2.06 for Windows, to work from Dos. It needs to be chainloaded, but it is usable with just dos and Hxdos (no mbr install).