Session hosts in a farm are redirecting connections not targeted to the farm's name

In a small Remote Desktop Server farm (2x 2008 R2 session hosts + 1 session broker with IP address redirection) of a large-ish domain, we noticed a change in behavior which is giving us headaches:

Previously, the RD session broker only did balance / reassign connection attempts to existing sessions when we were connecting to the RD farm's name ("rdfarm-generic"). Connecting to a specific session host by its DNS name or IP address did ensure that you were actually connecting to this specific host.

For an as of yet unknown reason, this has changed so that connecting to a session host in any possible way would get users reconnected to an existing session (if any). So a simple mstsc /v:192.168.21.85 (192.168.21.85 being the IP address of RDSH-02) gets me redirected to the other member of the farm if I have a session there already. This is what the series of events on the connection broker looks like in such a case:

TerminalServices-SessionBroker/800

RD Connection Broker received connection request for user AD\the-wabbit.
Hints in the RDP file (TSV URL) = tsv://MS Terminal Services Plugin.1.rdfarm-generic
Initial Application =
Call came from Redirector Server = rdsh-02.ad.example.com

TerminalServices-SessionBroker/803

RD Connection Broker has successfully determined the end point for this connection request.
Endpoint name = rdfarm-generic
Endpoint type = Farm

TerminalServices-SessionBroker/807

RD Connection Broker has found a disconnected session for user (AD\the-wabbit) with session ID (9) on end point (rdsh-01.ad.example.com). Load balancing will be skipped.

TerminalServices-SessionBroker/801

RD Connection Broker successfully processed the connection request for user AD\the-wabbit. Redirection info:
Target Name = RDSH-01
Target IP Address = 192.168.21.82
Target Netbios = RDSH-01
Target FQDN = rdsh-01.ad.example.com

Unfortunately, I cannot compare to older logs from regular backups - due to a configuration problem they were rotating away too soon to hold the users' connect entries logged early in the morning. So all I have is the disconnect entries from afternoons.

Why is it happening? How do I get the old behavior back?

FWIW, using mstsc /admin /v:192.168.21.85 does connect me to the destination server for sure, but obviously is unsuitable for users.