The only difference is how you access the windows data: from a partition or from a parallels file containing the windows OS. Windows thinks it's running on a bare machine. Windows, when run in a parallels vm, will present disk addresses to macOS like it would on a bare machine. MacOS will have to take these addresses and read the data for the vm.
(1) I have no idea how Parallels sees the partition. Let's assume there is some way for parallels to get direct access to the partition. Thus parallels would not need to translate the address into reading from a file. It could pass the addresses directly to macOS. MacOS would do some verification on the address and read the data and pass back to parallels.
(2) when windows is stored on a file. Parallels would have to translate the windows disk access into into a file offset. I don't this would take much processing. It turning a disk access into a file offset: basically two or three multiples and three adds. It passes this file offset to macOS. MacOS converts the files offset to a disk access.
The biggest will be if the data is stored on a rotating drive or a SSD. A rotating drive could result in a lot of disk arm movement because windows will be looking at the drive differently than how macOS has stored the data. This could be mitigated by some disk fragmentation software.
Anyway, having the data in a file as apposed to a partition could result in more processing cycles spent when access data from a file. The biggest factor still will be the speed of the drive.
Others have reported it doesn't make much difference. I've explored why that is so.
R