The EaganMatrix (inside the Osmose) and the Hydrasynth are both great and each one has its own approach. I think the Anyma synths are less beefy, in terms of computational resources, but the synth engine offers more kinds of modules, more freedom in some way. Not that it's always useful to have 16 LFOs or envelopes, or to be able to modulate the curve of a mapping, but it sometimes makes trying an idea easier during sound design. As we started with a wind instrument (Sylphyo), we also take special care to make support for this kind of MIDI controllers effortless.
The synth engine in the Anyma Phi runs on a STM32F4. The UI and MIDI routing runs on a separate STM32F4. No RTOS, we find it much easier to reason with cooperative multitasking, and easier to debug. So far, we don't have any latency/jitter issue with this approach, although it required writing some things (e.g. graphics) in a specific way.
The Omega runs on a mix of Cortex-A7 and STM32.
I have a pure software background but I came to appreciate the stability, predictability and simplicity of embedded development: you have a single runtime environment to master and you can use it fully, a Makefile is enough, and you have to be so careful with third-party code that you generally know how everything works from end to end. The really annoying downside is the total amount of hair lost chasing bugs where it's hard to know whether the hardware or the software is at fault.
In contrast, programming a cross-platform GUI is sometimes hell, and a VST has to deal with much more different configurations than a hardware synth, you're never sure of the assumptions you can make. The first version of Anyma V crashed for many people but we never had the case on the dozen machines we tested it on.