Thanks for the insight. I think I've figured it out.
First, the master sends a command to read 6 pieces of data containing the accelerations on 3 axis from the slave. Those are the first 7 clock bursts (1 command & 6 reads). Then the SPI bus is closed (CS goes high). Then it is re-opened (CS goes low again) and master asks for 1 more piece of data, what the current acceleration range is set to, then it closes the bus again. Those are the last 2 pulses.
Some math is done to the data in between these, and the data is always changing, so I think that's why the pulses aren't always perfectly in sync.
Basically what you were saying. Thanks a lot