These topics ARE troll magnets.
Namelos measured latencies for some keyboards, which sparked a good discussion. The link above in dmbr's post is dead, but the wayback machine does have that page. If anyone works out how to get page 9 and on from it, speak up!
Short version... all the latencies add together. Reduce any of them and it's a win. Namenlos' results showed that controllers varied a lot, but even the fastest could be improved on.
By polling rate, do you mean scanning the keys, or the USB polling rate?
USB polling is in most cases 125Hz, otherwise the advertising usually makes a point of it being faster (even then, some lie). Tools such as USBlyzer can query it. Low speed USB + any interval -> 125Hz, lol. High speed USB is the only way to get faster polling without hacking the OS.
Scanning rates are probably typically a bit higher now than they used to be, but still mostly around 200 to 500 Hz, at a guess. There's no easy way to measure it. Unless you have a scope or frequency counter - then it's easy! The only way to increase it would be to replace the controller. Faster scanning rates are good for two reasons - reduced latency, and reduced debounce time (indirectly, by being a better measure of time).