We're just coming to the end of a period of testing a variety of GPS receivers, with EGNOS both enabled and disabled. These varied from the latest models to (relatively) geriatric 10 year old models, and included hand held, chart table, and smart antennas types.
The intention if these tests has been two fold: to know what level of accuracy is achieved, so we can allow for this in our data processing, and, for the GSA, to measure what benefits EGNOS has on accuracy. In the tests, we logged the position of a receiver over 12 hours in a static location. For most receivers we had to use the vertical accuracy, as the horizontal position gets fixed to avoid random noise appearing on the display.
Although results are not yet finalised, they show that in general the vertical rms1 error was about 5m - this means that 69% of GPS fixes will have an error that is less than this. Roughly speaking, vertical errors are twice the horizontal ones for GPS, so this corresponds to less than 3m horizontal error. More surprisingly, we couldn't really notice any difference between when EGNOS was used and when it was disabled. As EGNOS corrects for the ionospheric and tropospheric errors, this suggests that errors are coming from other sources such as multi-path, receiver noise, clock and ephemeric errors. We had expected EGNOS to have a noticeably beneficial effect on the accuracy of the position fixes.
For more information, there will be a technical report downloadable from this site, and there will also be an article in Practical Boat Owner at a less academic level, that also discusses other factors.