It’s weird that reads should degrade. What benchmark programme are you using? For a quick and accurate benchmark I use AS SSD Benchmark.
Are you running AHCI or IDE?
My guess: assuming the benchmark writes a file to the drive first, and then reads it back to measure IOPs, then the file may be more optimally placed when written to an empty disk compared to written to a used disk. I say this because my understanding is that the flash is connected to the controller via multiple channels and performance is somewhat dependant on the number channels that can be used simultaneously.
Access time should be more or less irrelevant to where the data has been written. My guess is that the benchmark is giving a false reading when the drive is fresh.
There are some good benchmarks to compare performance here: