All in how the program is designed. The program decides how much memory to use, lets say it allocates 512Mb of the 4Gb total of physical address space available. 256 on card and 256main memory. It is then up to the kernel to handle that request, if there is enough available main memory then that becomes tagged as in use. There is a point here as to how the program is written, if it was written with a NT kernel compliance in mind, it should never try to read or write out of it's allocated address range. If not there is the possiblity during gameplay to have a BSOD due to the process trying to read or write to address block that it doesn't own.
Back to the allocated memory, if that is all the game made a request for and was allocated then that is it, more RAM will not change anything. If it uses only 100Mb of the 256Mb allocated during gameplay the other 156Mb block is still shown as being used, as it is locked from other proccesses.
If a system has less available main memory then it is up to windows as to what to page in and out of main memory to the disk to increase performance. FIFO kicks in, and starts offloading unused process modules and remapping their address space. This applies anytime the the system memory manager runs out of main memory. If a file shows recent enough use and the memory is still needed it will be paged back to the disk in sequential order for a faster load then out of the file system hiarchy.
So draw a pie and quarter it, one quarter is Firefox, one quarter is WMP, one is Messenger, and one is MS Word. Here we have all the system resources tied up. Now we quit everythign and go to do a little video editing that will use all the resources again. We have one of two possibilities.
1) The information in main memory is written to the pagefile on the disk as we have a high chance of using it again shortly, based on history useage, and how recently it was used.
2) There is no pagefile and to reflect last access time and date, as well as any changes made windows must rewrite all the active files to the disk in the file system, and depending on the level of fragmentation, as well as number of files this could take some time. If we had a pagefile all these changes would be taken care of at shutdown or when the memory manager offloaded the files from the active list.
So in all reality there is no scenerio where it is faster to not have a pagefile for the whole system to use. There might be one or two apps that run a bit quicker, and shutdowns may take less time, but the whole system will not be magicly faster.