Actually, while I don't think the AWE32 was designed with EDO RAM in mind, it's not that surprising that it works with EDO RAM. EDO and FPM only behave significantly different if you are actually keeping pages open for more than a single access or a single burst. Controllers designed for FPM RAM make EDO RAM cause bus conflicts when a page is left open speculating for further page hits.
A sample playback chip like the EMU8000 might use page mode to fetch four consecutive data samples for the four-point interpolation - but it doesn't use the data bus for anything else while it is fetching the samples, so no chance for a bus conflict. As the EMU8000 processes data in real time, it needs to use timings that work if every of the nominal 32 voices is played from a different page, so there needs to be enough time to close the page used by one sample before the page containing the next sample is opened. There is no potential performance gain that could be realized from page hits - we don't need to be faster than real time. Thus the most simple and straightforward implementation closes the current memory page after each individual voice is accessed, so there will be no conflict between a RAM voice and a ROM voice (which could happen with EDO RAM, if a page were left open).
