I actually store the phone number as a double precision. I decided to do
this in my own database since i beleive it is smaller and easier to work
with than a varchar(10). I don't have a client layer where phone numbers
are entered directly, but I do have a function (client side) that strips out
all characters except numbers. The numbers get entered in batches of 100K
to 1M at a time (My databases exceed 250 million numbers at the moment)
through a store procedure that checks to see if it is an actual phone
number. To perform the check it just casts the passed varchar(10) as a
double precision and then checks to see if it is >=1000000000 and
<=9999999999.

On the client side I find it rather easy to display the number with
formatting masks or "ugly" substring/concats.