I don't know where you read about USB interfaces needing to "compress the sound", but it's nonsense. For your application two channels will do; at 96 kHz and 24 Bit that's less than 5 MBit/s: even USB-1.0 can handle that without compression. and USB-2.0 is already more than ten times faster, so even in multichannel applications it's often plenty good enough.
What's taken as crucial for the performance of an audio interface is not so much bandwidth as latency, in particular when you're dealing with virtual instruments. FireWire has a design advantage here, not because of its higher data rate but because its topology allows devices to interact directly, and it can pretty much completely be implemented in hardware, with very reliable DMA performance. Whereas in USB there's a single root hub that needs to do some nontrivial packet management – not an issue for total bandwidth (just use big enough packages!), but it prevents reliable fast direct-memory access.
And this hasn't really changed with USB-3.0: it offers much better bandwidth than 2.0 (which however, as I said, is relevant only if you need a great lot of tracks) but is still worse than the old FireWire in terms of low-latency reliability, and much worse than Thunderbolt.
Again for your application, when you're recording from the amp with a microphone you don't need monitoring through software anyway, so latency doesn't matter either. So there's really no reason why you shouldn't be fine with just a USB-2.0 interface.