When a user checks in a document whose file was obtained from a local server's cache, the update sequence of the file data differs, depending on whether or not DFT is enabled.
If DFT is enabled, then the client will communicate directly with the remote storage area, and will synchronize the changes in the local copy directly to the master copy, effectively bypassing the gateway cache. This produces a measurably faster check-in process, at the expense of making the cache copy immediately stale. The next user to download the file from the cache then will pay a small performance penalty while the cache copy is synchronized to the newly-updated master copy.
If DFT is not enabled, then on check-in the entire file is transferred to the remote storage server via the gateway server. As the file data passes through the gateway server the cache is updated. Again, note that this can be quite inefficient when dealing with large files that have only accumulated a relatively small number of changes. However, a (small) benefit is that the cache is immediately updated with the current contents of the file.
Note that if DFT is not enabled, and multiple users in multiple locations are actively modifying a given file, the non-DFT cache update during check-in may provide very little benefit; it may very well be that the cache copy will be invalidated relatively soon by changes from another user in another location, and thus a full file transfer to the cache will be required on the next file download. Thus, the use of DFT is encourage for nearly all file operations; it is the best general-case solution to the problem of keeping distributed files and caches synchronized with minimal network overhead.