Once again I agree with Nigel. The cap does nothing, the leakage on the PIC input pin never "switches off". It causes an IR voltage drop on the source impedance all the time, so a cap changes nothing at all.
I know. I had the exact same problem, thought of the cap, and quickly evaluated that it has no useful effect in any configuration I can think of.
I do hate op-amp stages for this, the offset error can be large depending on the amp you use, in addition to limiting factors that the input probably can't go rail-to-rail anymore. So consider this. The 2.5K is the max input impedance to guarantee the PIC's ADC is accurate. If you exceed it, you can expect one +/- 1, or more, in additional error, but otherwise it does still work just fine. Check the spec sheet for the exact error versus input resistance. The thing is, if you only plan to use an amp with a 15mV offset, that's already +/- 3 codes of error. Depending on how much resistance above 2.5k you're trying to eliminate, it could actually be WORSE than without the buffering.