If I recall correctly, the idea that the path part of the URL maps to a filesystem path is an implementation detail, not covered in the protocol. So it might not map to a filesystem at all; it could map to an LDAP directory or some transient data structure in memory or it might not exist at all. There's no requirement that there even be a path part to the URL.
However I believe that the standard specifies that "/" (and not, say, "\") be used to separate the transfer type from the server&port and the server&port from the local identifier (which may use additional "/" characters).
I believe there's an RFC giving the standard.Use Google to find it.