There are always risks of exploit when allowing users to upload files, so it's good to be worried.
You say "I do some operation for the conversion", so while I cannot comment specifically on the safety of this operation, there could be risks here depending on the operation and the content provided to it (e.g. buffer overruns, invalid data).
I'm assuming you are saving your file with a .gz extension.
Saving with the client filename could pose compatibility problems if you do not clean the filename at all. There are differences in disallowed (or problem) characters between filesystems, such as & in Unix or : on Windows. Sometimes if you simply save a file, and simply read it again your code won't "find" it, unless you escape or strip all these characters properly.
The client filename could possibly pose a risk, if the name could have for example a path embedded into it. A presented filename such as "../../../../home/user/file" could possibly trick your program into overwriting a file, as long as the permissions were very badly implemented and you are simply concatenating paths. At worst I'd say this would be an annoyance or DoS attack - limited to overwriting gzip files and "breaking" them.
The client filename could possibly overwrite another user's files? I'm not sure what your namespacing is, but a clever attacker could try to trick another user into downloading their xml/sql file by naming it cleverly.
Also if you could guess someone else's filename, you could guess the resulting URLs and war-dial through them looking for content.
All of these risks would go away, if you used a nice GUID to name the file. Or mapped it to each user's session (e.g. file1.gz is only valid to that user's session).
I generally don't use client names, or seriously validate and clean them before re-presenting them.