90 nm (.09 micron) usually refers to the gate length. As you are probably aware, a transistor has three parts: a source, a gate, and a drain. So the actual size of the transistor will be more than 90 nm. And with CMOS transistors (which make up most of the transistors in modern CPUs) there are actually two transistors: a pMOS and an nMOS.
What smaller transistors actually mean, as others have said, is: a) reduced power, b) increased switching speed, and c) higher circuit density.
CMOS transistors have an intrinsic capacitance. Making them smaller reduces this capacitance, so they require less current to operate. They also require less voltage to turn "on". Both of these reduce power.
Smaller transistors can switch faster (although size is certainly not the only factor). And since they're closer together, the propagation delay along the metal lines and vias can be reduced.
There are trade-offs though. When CMOS transistors switch faster, that of course means they switch more often. So more charge might be dissipated per second, even though the charge per switch has been reduced. Smaller transistors also tend to be "leakier", meaning more current flows through the transistors even when they're not switching. Making transistors less leaky slows them down though, so it's a very delicate balancing game for CPU designers.