SSE2 is great for more than just SIMD. It gives you additional registers, allowing the compiler to generate more parallel scalar operations. Visual C++ with /arch:SSE2 will happily interleave x87 instructions with SSE2 instructions, which looks strange but works well.

You may also want to use SSE2 for saturated scalar additions or prefetches.

Very few people have x86 processors that don't support SSE2 these days. In our tests at IMVU, I think less than 1% of our customers didn't support SSE2, and it was only that high due to the Athlon XP.

In short: when starting a project targeting x86 today, default SSE2 on.