The 550Ti should be around 15% faster than the GTS450.
While you have only run two NATHAN_CB1 tasks on your new card, it looks like the GTX 550Ti is ~7% slower, for that one task type.
My guess is that other activity is interfering with GPU crunching. Looking at your results it appears that you stop crunching when the computer is in use. This slows down crunching and would detract from the accuracy of comparing tasks. I suggest you configure Boinc to 'use GPU when computer is in use'. Also of note is what the CPU is used for. Saturating the CPU with CPU tasks can starve GPU tasks of essential CPU time.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Two relevant posts:
FAQ - Recommended GPUs for GPUGrid crunching
Relative Comparison of Recommended Cards

A few things perhaps worth checking,
Did you increase the number of CPU threads being used?
Did you start crunching more demanding CPU projects, say with high HDD I/O, or start optimizing for CPU projects?
Were you overclocking the old card?
Did you change CPU clocks?
Were you using SWAN_SYNC and then removed it?
Per chance did you put the 550Ti in a secondary PCIE slot - an X8 or X4 slot?

If none of the above (or similar) I wonder if anything was 'trimmed down' on the 550Ti that would make it perform below the expected (based only on GFlops peak and Compute Capability). It's possible that the GF116 cards might be relatively much worse than the GF114 or GF104 cards. I would categorize the GF116 cards as low end, as they are only CC2.1 (only use 128 of the 196 shaders).
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help