Obviously, we wanted to test if ReadyBoost has a measurable impact on system responsiveness as Microsoft claims. If indeed it does work, then it should be a versatile and relatively cheap way to boost your system performance. In order to do so, we set out to create a repeatable scenario where we could measure to a certain degree of accuracy, the effect of having a ReadyBoost cache in Windows Vista. Our trial and error finally yielded a scenario where the user is trying to close four applications, in our case, Firefox 2.0, Open Office 2.1, Windows Mail and Paint.NET (opened with 40 images).
For all the tests, we used the same Sandisk Cruzer 4GB Titanium thumb drive and manually set the amount of memory (e.g. 512MB or 1GB, etc) to be allocated for ReadyBoost in the Properties dialog box of the drive for each instance. Since ReadyBoost was supposed to enhance system responsiveness, we then timed how long it took to close these four applications and here are the results:
From taking almost a minute to close the four applications without the benefit of ReadyBoost, we got this timing down to around 23 seconds. That was with a 4GB thumb drive. Even then, a more modest amount like 512MB of flash memory immediately saw a steep drop in the time taken. The returns however start to diminish at 2GB, so perhaps that should be the effective maximum cache size for this system. Microsoft itself recommends a 1:1 ratio between system and flash memory, especially for a slower system and the greater payoff (31%) for 512MB of ReadyBoost does seem to confirm that rule of thumb.
For the mid-range system, there were obvious gains with ReadyBoost. This time, we saw a rather consistent improvement in results going from without ReadyBoost to having 512MB and then going to 1GB. Such an improvement seems to suggest that the 'sweet spot' for this system is closer to 1GB and perhaps even 2GB. There were hardly any improvements once we hit 2GB and while we did record slower times for the 4GB test, it was a negligible difference of around a second.
We tried to duplicate this same set of tests for our high-end system but the results were not conclusive so we have not shown them here. No matter what amount of ReadyBoost cache that we specified, there were no discernible time savings. Most likely, the system was already so fast (it took around 7 to 8 seconds to complete the test) that the ReadyBoost was having minimal impact on performance and therefore the difference could not be measured with our 'primitive' stopwatch. No doubt, there probably is some extremely tough workload out there that could eventually stress the system and utilize the ReadyBoost cache but considering the average user, it is unlikely to happen often.