To put it simply, it doesn't do the same thing as older Bluetooth chips.
The new chip uses BLE, or Bluetooth Low Energy. This is a mode which transmits packets at a high bandwidth in very short bursts, enabling the exchange of data with a very low duty cycle. When transmitting or receiving, power consumption is comparable with standard Bluetooth, but the time that this actually happens is so short that it doesn't really matter.
The laws of physics haven't been changed, instead, the protocol has been changed (or, more accurately, a new protocol has been added to the Bluetooth spec.) I'm not sure what trade magazines you're reading, but this isn't the only chip to implement this protocol. It's a standard protocol, and TI and Nordic Semi also have implementations. Instead, 'Bluetooth' in this context doesn't mean what it used to. This isn't a bad thing, but if the magazines confused you to think that Broadcom (and Broadcom only) has made a huge breakthrough in Bluetooth technology, I'd re-evaluate my magazine selection.
I've been interested in giving it a try, and have been looking at Panasonic's PAN1326 modules as a start. They integrate a chip antenna with a filter/LNA and a TI CC2564 under an EMI shield in a little module (about 1 cm\$^2\$ - you might want to wait for a breakout board if you're uncomfortable with soldering QFNs or can't do a PCB layout). The CC2564 requires a 32.768 kHz clock input and power (1.7-4.8V), and provides a UART and I2S interface. It's available for engineering samples and in small quantities from Mouser and Digikey. [Sorry, that reads like an ad - no affiliation. Just a summary of the specs, check the full datasheet here]
No, they won't work. CR cells are designed to deliver maximum a few mA continuously, so a low power microcontroller like an MSP430, for instance running off a 32.768 kHz crystal, may work, but the power for an RF transceiver is much too high.
The datasheet for this battery even specifies only 0.2 mA as "continuous standard load".
If you would need your RF in short bursts then you could get the power from a large electrolytic capacitor, for say 100 ms, and then recharge it over a longer time, so that the controller still gets enough current. For continuous streaming this won't do.
Best Answer
I believe that the wireless mice and keyboards only transmit when an event happens, while BLE has an ongoing "pairing" traffic. TX-on-demand moves the power consumption to the reciever (which is not on battery) but it makes it impossible to tell when the keyboard has gone out of range.
They're also single-direction (apart from keyboard LEDs, which may not be present), while BLE can exchange data in both directions.