It might happen the MIM cap is double counted - make sure that is not the case.
By saying double counted I mean - it was already extracted during LVs netlisting, then RC extraction tool again extracted the cap fully - because of lack of blocking statements [maybe]
Check-A
----------
check the QRC/RCX log file.
There will be a section showing [maybe Forking: ] capgen -[several_options_especially_check_for_blocking_options] - you must see the MIMcap_device_recognition_layer and a comma separated list of layers that is blocked from cap extraction. If you cannot find so, definitely a double counting has taken place.
1. check every blocking statement.
2. be careful, the device being blocked may have a name *not* same as device/cellName, it is it's corresponding device recognition layer name [you can find the link by taking a look inside the extraction rules used during LVS].
Check-B [ you can do Check-B, b4 Check-A ]
------------
Another way to check - check your LVS layout netlist - check for total value of all of canonical caps for the net, where one or,maybe couple of MIMcaps are instantiated , now re-run RC extraction with a simple spice netlist out with decoupled cap only extraction [this time you may skip R extraction], if extracted netlist, total cap on the net, is more than 2x[LVS value] - it is perhaps double counted during extraction.