performance regression when groupby acts on multiple columns

For example, we have a frame which the shape of it is(34284584, 18),
when by only acts on one column or a few columns, like:

`df[:, {"A":min(f.A), by("B")}] or df[:, {"A":min(f.A), by(["B", "C"])}]`

It takes about seconds.

But when by acts on multiple columns, the time-consuming increases to the minute level, like:

`df[:, {"A":min(f.A), by(["B", "C", "D", "E", "F", "H", "G", "J", "K", "M")}]`

Performance is even weaker than single-threaded pandas.

Is there a performance problem with groupby based on multiple columns?
Thanks a lot!