I2C has two critical limits that must be respected for the bus to work. The current that any device on the bus must sink on SDA or SCL to pull it down to logic '0' and the risetime of the SDA and SCL signals, which needs to be significantly less than 1/4 the bit rate to guarantee SDA is stable at the SCL referenced bit sampling time points. The least of the max logic '0' sinking currents for all devices on the bus, and the max supply voltage sets the lowest value of total (paralleled) pullup resistance you can use, and that in turn multiplied by the total bus capacitance determines the rise time and thus the fastest bit rate the bus can run at. The bus capacitance is highly dependent on board layout and interconnects, but is usually the sum of a capacitance proportional to the total bus length plus a capacitance proportional to the number of devices on the bus.
The details and necessary calculations can be found in sections 6 & 7 of the
I2C-bus specification and user manualYour daisychain configuration is preferable if you need to minimize bus capacitance.
3x 1K2 resistors in parallel is only 400 ohms, well out of spec for any I2C devices not capable of fast mode plus (20mA sink current) operation. As Bassman points out, you only need a single pullup, and if there is a single bus master, put it there. In the case of a multimaster bus, you still only *NEED* a single pullup, but splitting it so there is a pullup at each master may be preferable for reliable bus fault detection.