Hi
(1) may be most reliable speed-wise, if changing hardware is no-no.
(2) CC256x has totally different package and, if I am not mistaking, needs host MCU to make use of Bluetooth via HCI interface (256x provides UART for this). CC2540 could be used that way, too (with host sending GATT/GAP or lower level HCI commands via UART). Still, their packages/pinouts are not same.
If you have CC2540 as a single device (no host MCU), then CC256x is probably out of question.
A system using CC2560 as a BT 2.1 endpoint will most certainly allow for more than 10kbyte/s. As of connection procedure - well, that's up to you to decide how the do it, but with SPP, there still will be initiator for SPP link, and there will still be a device passively waiting. I don't quite understand how the y could do it independently.
For starters, just get TI's BLE stack at this wiki page. Do you have working hardware/development kit at hand? Their stack installation contains two staple examples of BLE Master and Slave devices (Central and Peripheral roles as they call it), they are good to start your own application with just minor changes.
Also, there is semi-useful MasterSlaveSwitch example, which contains very basic code for switching between Peripheral and Central devices.
I do recommend to at least skim through so-called Bluetooth Core document. It is frighteningly large at first glance, but you don't need it all to start - just book 1 and book 6 (the latter is about BLE).
BR,
Oleg
