Zenoss is quite powerful, but the default installation cannot sustain hundreds of devices very efficiently.  These tweaks might help.

1. If necessary configure distributed collectors.

2. Increase the number of running zenhub threads:

$ZENHOME/etc/zenhub.conf

# 3 workers has been proven in real-world systems with 4 cores
# but experimentation is necessary to get a value for your own
workers 3
# increase zenhub memory cache
cachesize 8000
pcachesize 2250

3. Increase ZOPE pool-size:

$ZENHOME/etc/zope.conf

# BEWARE:  increasing zserver-threads might lower the
# overall performance of Zenoss
<zodb_db main>
  mount-point /
  # ZODB cache, in number of objects
  cache-size 60000
  pool-size 50
  ....
</zodb>

4. Decrease concurrent running zencommand:

$ZENHOME/etc/zencommand.conf

# reducing how many concurrent commands zencommand runs will consume
# fewer resources at the cost of longer finish times.
parallel        5

5. Decrease concurrent running zenmodeler:

$ZENHOME/etc/zenmodeler.conf

# reducing concurrent modeling jobs will consume fewer resources
# at the cost of longer overall modeling times
parallel        5

6. Decrease concurrent running zenstatus:

$ZENHOME/etc/zenstatus.conf

# reducing concurrent checks for status will consume fewer
# resources at the cost of longer cycle times
parallel        5

7. Tweak mysql memory usage:

$ZENHOME/mysql/my.cnf

[mysqld]
table_cache = 1024
thread_cache = 16
query_cache_size = 64M
innodb_buffer_pool_size = 512M
innodb_additional_mem_pool_size = 50M
innodb_flush_log_at_trx_commit = 2
innodb_thread_concurrency = 16

For a more thorough discussion of mysql tuning consult the Reference.

Although this may help with mysql performance for the events database, disk I/O is still a major factor in overall performance of Zenoss.  You will need good throughput if you are monitoring a large number of devices.

Zenoss makes use of RRD for storing and graphing data.  Learn about RRD Tuning.