Here are most of the BLE chips I know:
TI CC254x: old, well known, power efficiency is not great, extremely widely used. power: 27mA@any supply voltage, size: 6x6mm, price: $1.95. NOTE: no DC-DC converter on board.
nRF51822: newer, excellent, good if a non-realtime MCU is needed, power: 8.0mA@3V, size: 3.5x3.8mm (more commonly 6x6mm), price: $1.92.
nRF8001: newer, excellent, good if you don't want to develop firmware; essentially same as above
Broadcom BCM20732: very new, good but support is lame, also good if non-realtime MCU is needed. power: 9.1mA (?), size: 6.5x6.5mm, price: $3.95. NOTE: this is a system in package (SIP), it only needs an external crystal - balun and antenna are on board.
ST BlueNRG: very new, looks good from datasheet, power: 8.2mA@3V, size: 2.66x2.56mm, price: $1.45.
Dialog DA14580: very new, looks excellent from datasheet. power: 4.9mA@3V, size: 2.5x2.5mm, price: $1.68. NOTE: this includes a balun on board, still needs crystal and antenna.
Important questions you need to think about:
Do you need FCC approval? Do you know how to go about it? (If not, use a module which has approval already)
Do you know how to tune the RF part of this? That needs special tools (network analyzer, etc) and expertise. When I had to do that, it was a real pain :) (If not, use a module, or use the Broadcom chip which has antenna on board; or Dialog chip which integrates a matching network and has a reference design for antenna)
Do you want to develop firmware for the Bluetooth SoC? That can be complicated/annoying/expensive. (If not, use nRF8001 or BlueNRG, or use TI CC254x which has ready-made firmwares already)
What is the primary design driver/goal: size, unit cost, development cost, time-to-market?
Considering the different possible goals:
For total size, the Dialog chip is hard to beat.
For unit cost - I'm not quite sure, depends on your manufacturing capability, but don't ignore the fact that some of these chips need way more support components than others (so that $3 Nordic chip is really more like $5-$6 per unit for the complete subsystem, whereas CC2540 based modules can be had for less than $5). Also don't ignore the $10-$20k (?) for FCC approval amortized over however many units.
For development cost or time to market - do yourself a favor and use a module :)
On your other questions:
I've already got a micro on the board so I don't really need another SoC. A BLE radio would suffice.
It seems that nRF8001 and nRF51822 (for example) only differ in that the firmware is burned in (not upgradeable, fixed function) on nRF8001; almost everything else about these chips is the same, leading to the supposition that they are built on either exactly the same, or almost the same, silicon. So there is no savings in power, part count, etc; there may be savings in ease of development (don't need to use the BLE stack, can use commands to control the device from another MCU).
Given that iBeacons are just transmitters, are there any BLE radios that do not have the RX capability (which would hopefully translate to lower costs)?
No, sorry; many things in the BLE standard require two-way communication.