No, this is fundamentally something that most file systems don't support. (There may be some with special data structures to make it cheap, but most don't - and the standard APIs don't support the idea.)
The simplest way of "deleting" part of a file is to copy just the bits you want to a new file. If you're only removing bits near the end, you could potentially avoid copying the start of it again, but that would require more complicated code.
In the case of a stream of objects, you'd probably be best to just read objects from one stream, and write the ones you still want into a new one. (Obviously you can rename the output file afterwards.) It's possible that by looking at the serialization format yourself you could avoid some of the processing required by repeated deserialization/serialization, but I'd go for the simple route to start with.