Citrix

High Availability can mean a lot of things and you can also go over board with HA, but that depends on how critical and how much money you're willing to throw at it.
Are you talking Local HA (i.e. making sure each XenApp component in one location can't cause an outage)?
Are you talking about Geographic HA (i.e. making sure if a location completely fails you can bring it up at the second site)?
I'll break it down into components for a single location.
1. Web Interface: I'll assume you're using WI for Web or PNAgent.  You'll need at least 2 WIs.  These servers will need to sit behind a set of load balancers (Netscaler, F5, Cisco)  in Active/Passive mode or Active/Active mode (not sure about Active/Active since I'm not a networking guy).
a. This way if a load balancer fails or if a WI fails it will fail over to the secondary.
2. XML Brokers:  The WIs use these servers to authenticate users, enumerate what applications the user has access to, and direct them to the appropriate server (ex. typically the server with the least load that hosts the application).
a. These servers can be dedicated, your ZDC(s) (If you have dedicated ZDCs and a Most Preferred and Preferred), or servers hosting applications or desktops.
b. As to what option in "a", I  would say with the size of your environment you don't need dedicated XML brokers and can use the ZDC(s).  You'll want to use two obviously for redundancy.
c. You can even load balance the XML brokers with a load balancer, but I would consider this extreme even for a large environment (100's of servers/1000's of users).
3. ZDC servers:  Discussed in "#2", but with an environment your size I would say you specify an Most Preferred and Preferred ZDC.  Now these can be dedicated or they can be servers hosting applications or desktops.  You could probably get away with using servers hosting applications or desktops that have low utilization, but this is hard to say without knowing your environment in detail (server specs, users per server, etc.)
4. Datastore: I would say for an environment your size you do not need to use any SQL redundancy.  If a Datastore goes down, the farm will work indefinitely but while it is down you can't make any changes to the farm.  You could implement SQL HA, but it may be overkill.
5. License Server:  Just one.  Even if your license server goes down you have 30 days to get it back up.
Now you say that you have 4 farms. You'll need to consider "#2" and "#3" for each farm.  As for "4" you could have all the datastores on the same SQL server, but again if you lose that server you can't make any modifications to any farm till it is back online.
Another topic of discussion, but if business requirements allow you could consolidate the servers into one farm.

Oh yeah, one more very important question:  will you be using physical servers, virtual servers or a combination of both?