In theory the one with the faster bus speed will be faster. With the faster bus the data can get to the CPU and get worked on faster. Reality says that both are 3.8GHz, so you'd need a benchmark to see any difference. I vote you use whichever one allows for the lower voltage/temps.
If you Decrease the bus speed you are reducing the "speed" at which your Memory, PCIe And PCIEx16 Functions on your system(among other things).
If you are running your PC at the Current 19x Multi Then i would strongly suggest against Going to the 22X multi, you will see FPS drops and rendering increases as well as other notable performance drops.
I agree that theoretically the higher base clock will be faster, but truth is I highly doubt there will be any noticeable difference except maybe the most extreme synthetic benchmark. Base clock is tied into Vtt/IMC voltage which I believe the i7 max is 1.35V. So you could get lower temps and power consumption at the same overall core speed. Or you could find a happy medium. Truth is tho each system is unique so if you want to find out what's best, you'll have to test it and run some benchmarks. IMO use Intel Burn Test as it gives you a speed output in gigaflops for comparison. A little math of figuring out how efficient you are:
In theory you can run 4Gflops per core per ghz. So if you're running hyper threading, you would have 4x8=32Gflops per ghz. Therefore at 3.8ghz, you can have 32x3.8=121.6Gflops. Actually truth is I don't know how hyper threading affects it, logically it should act as an extra core... I know the math works out like this with physical cores only for certain. ANYWAY! My point is, I ran a bunch of tests and found that certain speeds lowered the efficiency of how close it was to theoretical Gflops. IE: at a slower speed I was about 8% off theoretical while at a higher speed it's more like 10-11% off, although the actualy Gflops is still higher - just not as close to the max theoretical. I also noticed that simply changing vcore could affect Gflops a bit. There's often somewhat of a sweet spot that'll make it just a tad faster, and I noticed too that too much under or overvolting would throw it further out of whack.
TL;DR = test both with Intel Burn Test and compare the resulting GFlops. Make sure to turn off all background apps/tray tools/gadgets even end processes you don't need. Run as much memory as you can. This tests stability and speed.
Or you could download Everest and run all kinds of benches for both your processor and memory, but like already stated you won't see a big difference in day to day computing UNLESS you are running the Large WUs, then it may be worth the optimization. IF so, and your temps are OK stick with the higher Bclk.
Not surprising? By using HT the pipeline stays fuller. This means more stages in the core stay active, meaning the temps go up. What you should see assuming the software your running responds to cores is a reduction in time needed to do something.