This page contains a good explanation imho:
Since aliases are, in a sense,
duplicate tags, the system needs to
know whether the subnet represented by
the tag is new, and this is indicated
by the netmask. With a "normal"
netmask, the system is told this is a
new subnet, and it can then set up
internal tables correctly. If the
subnet is not new, the tables will get
set up incorrectly if the netmask is
"normal".
With a netmask of 255.255.255.255, the
system knows this is a duplicate of an
existing subnet, and therefore will
assign the address as if it were
assigned to the loopback interface,
with the "point-to-point" mask.
Assigning a netmask of 255.255.255.0
will, in most cases, appear to work.
However, the internal tables will not
be set up correctly, and if the alias
is removed, problems may ensue. This
can be overcome by adding appropriate
host routes when the alias is added,
and removing the routes when the alias
is removed. However, the system works
correctly with the "point-to-point"
netmask, and maintenance is easier.