Yup. the Synapse server implementation is still beta, and we are working away on the performance issues (yay, Python/Twisted). In the last few months we've reduced the memory footprint by about 2-3x (by string interning, better caching, and letting the cache size be configurable). The only stability issue we're aware of currently is the risk of being OOMed on small VPSes.

That said, we're really hoping that having almost finished fleshing out the whole ecosystem now, the future will be much tighter and svelter homeservers like Perceptes' ruma implementation.