The LDAP protocol allows for creating new extended operations or controls on already existing operations. This however involves quite a lot of work in defining the extensions and writing code (on both the server and client end) to implement it. You need deep knowledge of the inner workings of your ldap server to implement them and any bugs/problem in your code will take the whole server down. So it’s a solution that:
* Does not scale
* Is error prone
* Is platform specific (a plugin for Sun ONE will not work for OpenLDAP)
* Involves too much work implementing it
* Requires LDAP operations instead of using web services (which is a big plus if you are dealing with outside web based clients)

LUMS could theoretically be written as an LDAP server plugin but that would require writing it in C instead of PHP (which is a lot more work) and creating one version for each LDAP server available on the market (or maybe just creating a back-perl backend for OpenLDAP in order to save time). In general, too much work, too much risk with minimum benefit.

Is the introduction of yet another layer like web services for something that looks so obvious/common necessary? Is the protocol and the implementations so poor in features?