Not only do both antennas have to be the same model, but they both have to be oriented to cover the same area.
Here's an explanation from Cisco's paper on multipath and diversity:
"A golf course with an electronic scoring application uses an access point with an outdoor antenna to cover an area of the golf course. One antenna is used to cover the left side of the course. Because there is little multipath, one antenna is sufficient. The course uses a directional Yagi antenna for its distance capability and ease of installation.
"When the golf course wants to add coverage to the right side of the course, the staff does not add another new access point to acheive this. Instead, it attaches a directional Yagi antenna to the other antenna connector, and points it in another direction. The staff drives around the golf course and performs a site survey to test the network. There are no issues with coverage. However, when tournament play starts and more users are added to the wireless network, they start to encounter difficulty and loss of connectivity.
"When the client on the left side of the course associates to the access point, it has very low signal strength, because the access point picks up the signal from the client on the right-pointing antenna. As a result, the client is out-of-range of the right antenna and drops its connection. However, the access point radio detects a problem and samples the left antenna port, under the assumption that it has encountered a multipath problem. The antenna switches over and the client increases coverage. As the client moves to the other side, retries begin and the access point radio switches over, uses the other antenna port, and preserves connectivity.
"Thus, when the access point cannot receive the client signal, it switches. The access point evaluates and uses the best antenna to receive client data. The access point then uses that same antenna when it transmits data back to the client. If the client does not respond on that antenna, the access point tries to send the data out the other antenna.
"In this scenario, the initial configuration was one client and two separate coverage cells; this works until additional clients are added. As the access point communicates to clients on the left side of the course, it does not switch to the right antenna port if no retries occur, because it does not detect any errors. However, it causes difficulties for users that are not on the left antenna.
"Note: The two antenna ports on the access point are designed for spatial diversity, and the radio only checks the other antenna when it encounters errors.
"The clients on the right side of the course have difficulty with connections. Only when a client with a weak signal reaches the left antenna does the access point recognize those clients and switch over to pick them up. This makes the right antenna active, so the left side of the course starts to receive errors until the antenna on the right hears a client from the left and switches over again."
