HTTP request sent to down member

I have several HTTP applicaitons hosted on my LTM running 10.2.1. On one of our pools, the custom HTTP monitors frequently fail on the members. They are legitimate failures, but it seems that requests are still sent to the down member after it is marked down. I think I'm missing some kind of configuration setting somewhere. Here is what I'm doing to test the behavior:

1. I constantly refresh a page on the virtual server that returns the server number. Since the pool is round robin and no one else is hitting it, it just alternates pool members for me.
2. I break the health check on 10.10.99.60 by stopping the Cold Fusion service and confirm the member is marked down within 21 seconds (interval + timeout).
3. When the member is marked down,every other request is still sent to 10.10.99.60. The server still responds, but with a 503 error. This still occurs several minutes after the member has been marked down.

Here are the relevant settings:

monitor http_200status {

defaults from http

recv "HTTP/1\.(0|1) 200"

send "GET /probe/heartbeat.cfm HTTP/1.0\r\nConnection: close\r\n\r\n"

pool Pool-B {

monitor all http_200status

members {

10.10.99.40:http {}

10.10.99.60:http {}

}

}

virtual VIP-1 {

snat automap

pool Pool-B

destination 172.16.0.1:http

ip protocol tcp

profiles {

tcp {}

}

}