28kbps is pretty slow. Just about anything could manage that.
Does 5 pins include power, or you have 5 signal pins available?
Full duplex UART would be only two pins, but would require that all devices agree on a baud rate. A clocked interface might be better, because if you later decide you need more throughput you won't have to go changing baud rates everywhere. It also means you could more easily get by without crystals on some of the sub boards. That means USART, SPI, and I2C are options. All will fit into five pins easily, but wouldn't leave you enough pins for discrete chip select lines, so you will have to have some sort of addressing mechanism built into your protocol. I2C has this by default, so that's a point in its favor, but depending on the overall system size and number of devices, it may be challenging to get it working at higher speeds. SPI or USART would be somewhat easier in this regard since they're actively driven from both ends. SPI has an advantage in that the CS signal provides an inherent frame delimiter, so you can use that to allow your slave devices to identify the beginning and end of a message easily. With a USART you'd have to include either a separate framing signal (at which point you basically have SPI), or define a packet format that includes some sort of frame delimiting, or use a break signal or timeout to delimit packets.
A one master/many slaves approach is definitely simpler. Multi-master systems are possible, but you have to have some sort of bus arbitration system to resolve situations where two masters try to take control of the bus at the same time. A single master system does require that the master poll all of the slaves for status changes, so it would be a good idea to include an interrupt line so that slaves can signal when they need the master's attention.