I have a big cube the I cannot put too much aggregations, otherwise I will
get memory erro: server run low on memory.... I know all dimension data will
be read in memory.

However, I found one of the table (200 rows) have a lot of columns, which
are used for levels of dimension, use char(100), char(50) as data type. The
actual strings in these columns are usually far less then the number defined.
If it's possible to use less memory by changing the char to varchar? I am not
sure if SQL Server will automatically save space when reading char(xxx) into
memory.

There are other dimension table that has thousands rows, but it already use
varchar.