Python Pandas qcut behavior with # of observations not divisible by # of bins

Suppose I had a pandas series of dollar values and wanted to discretize into 9 groups using qcut. The # of observations is not divisible by 9. SQL Server's ntile function has a standard approach for this case: it makes the first n out of 9 groups 1 observation larger than the remaining (9-n) groups.

I noticed in pandas that the assignment of which groups had x observations vs x + 1 observations seemed random. I tried to decipher the code in algos to figure out how the quantile function deals with this issue but could not figure it out.

I have three related questions:

1. Any pandas developers out there than can explain qcut's behavior? Is it random which groups get the larger number of observations?
2. Is there a way to force qcut to behave similarly to NTILE (i.e., first groups get x + 1 observations)?
3. If the answer to #2 is no, any ideas on a function that would behave like NTILE? (If this is a complicated endeavor, just an outline to your approach would be helpful.)

Here is an example of SQL Server's NTILE output.

Bin |# Observations
1   26
2   26
3   26
4   26
5   26
6   26
7   26
8   25
9   25

Here is pandas:

Bin |# Observations
1   26
2   26
3   26
4   25 (Why is this 25 vs others?)
5   26
6   26
7   25 (Why is this 25 vs others?)
8   26
9   26