To solve the poor bandwidth efficiency problem of VoIP over IEEE 802.11 wireless LANs (WLANs), the literature frequently suggests elongating voice payload size. Such "voice frame aggregation" can be performed (1) over a single call or multiple calls, (2) at the voice source or at the wireless access point (AP), and (3) with additional repacketization delay or without it. This paper shows that over IEEE 802.11 WLANs, voice frame aggregation should better be done at the AP, over a single call, and without additional delay. We demonstrate through simulation and analysis that this particular implementation choice enables us to support far more VoIP calls on 802.11 WLANs than other choices. For one instance, we show that 100 or more G.729 calls are possible on 802.11b within ITU-T recommended delay bounds.