The latest I read is 3 but that was in Oct. With Lene Hau of Harvard's "frozen light" and with quantum donuts, newer strategies for stabilization are appearing, but the problem of keeping the qubit in superposition for long enough to sample it seems boggling. http://www.sciencedaily.com/releases/2009/03/090309105026.htm
Hau has demonstrated a memory for classical light pulses (that should also work as a memory for quantum states of light) but her system is - as it stands now - not a viable architecture for a quantum computer. I'm not familiar with the work you link to from sciencedaily (which is not Hau's work).
I'm pretty sure the state of the art for number of qubits (in an architecture that is viable for building a quantum computer) is held by ion traps, with numbers approaching a dozen qubits. Two groups that are pushing the state-of-the-art are the NIST ion trap group and Rainer Blatt's group.
See, ferinstance, one of the latest papers from the Blatt group, with 14 qubits: http://arxiv.org/pdf/1009.6126